Learning from a single labeled face and a stream of unlabeled data
This work proposes Online Manifold Tracking (OMT) to recognize faces using only one labeled image and a continuous stream of unlabeled data. OMT learns the underlying structure of human faces dynamically, treating it as a non-parametric model. It achieves high accuracy by using nearest-neighbor classification on this learned manifold, outperforming methods that require more initial training examples.